基于VMD-SE和机器学习算法的短期风电功率多层级综合预测模型  

Short-Term Wind Power Multi-Leveled Combined Forecasting Model Based on Variational Mode Decomposition-Sample Entropy and Machine Learning Algorithms

作　　者：张亚超[1] 刘开培[1] 秦亮[1] 

机构地区：[1]武汉大学电气工程学院,湖北省武汉市430072

出　　处：《电网技术》2016年第5期1334-1340,共7页Power System Technology

基　　金：国家重点基础研究发展计划项目(973项目)(2012CB215101);国家自然科学基金项目(51309258)~~

摘　　要：针对风电功率受自然环境变化影响,难以建立精确数学模型对其进行预测的问题,采用一种新型的可变模式分解(variational mode decomposition,VMD)技术,将原始风电功率序列分解为一系列有限带宽子模式以降低其不稳定性,根据子模式的样本熵(sample entropy,SE)分析其复杂度并重组得到子序列。在此基础上,结合3种不同的机器学习基模型,提出一种基于VMD-SE和基模型的自适应多层级综合预测模型,并采用一种基于混沌萤火虫结合仿真鸡群优化的智能算法,对其权重矩阵进行实时调整。仿真结果表明,基于VMD的组合模型较采用聚类经验模式分解时预测精度明显提高,且所提综合模型的预测精度较组合模型有了进一步的改善。因此,所提综合模型能有效提高短期风电功率多步预测的准确性。In view of difficulty in establishing accurate mathematical model for wind power prediction influenced by environment changes,a novel variational mode decomposition（VMD） technique is adopted to decompose original wind power series into a set of band-limited sub-modes to decrease instability.Then sample entropy（SE） value for each sub-mode is used to analyze its complexity.The sub-modes can be recombined to obtain a set of subseries.On this basis,combined with three machine learning-based base models,a self-adaptive multi-leveled combined forecast model based on VMD-SE and the base models is proposed.Its weighting matrix is adjusted with a bio-inspired chicken swarm optimization（BCSO） algorithm integrated with chaotic firefly in real time.Simulation results demonstrate that VMD-based hybrid models perform better than hybrid models based on ensemble empirical mode decomposition（EEMD） technique.Moreover,the proposed combined forecast model can further improve prediction accuracy compared to above hybrid models.As a result,the proposed combined model can enhance prediction accuracy of short-term wind power multistep forecasting effectively.

关 键 词：短期风电功率多步预测 可变模式分解 机器学习 仿生鸡群优化 多层级综合模型 

分 类 号：TM715[电气工程—电力系统及自动化]